LunaTalk

The Challenge

• Problem: There are very few opportunities to practice natural and enjoyable Japanese conversations.
• Situation: Learners struggle to find affordable and flexible ways to engage in real Japanese dialogues.
• Necessary: An AI-powered platform that makes Japanese conversation fun, accessible, and personalized.

The Goal

• Objectives: Build an AI chat app that helps learners practice natural Japanese conversations. Provide personalized learning paths and vocabulary management to boost confidence and fluency.
• Metrics: Track progress through MRR, MAU, and learner's self-reported confidence.
• Impact: Enable users to practice Japanese conversations consistently, nurturing a lasting love for Japanese.

User Research

User Journey

Design & Development

To address learner's needs for natural Japanese conversation, quick vocabulary management, and easy onboarding, I defined core requirements and chose a tech stack that supports them.

Tech Stack & Tools

• Tech Stack: React Native(Expo), Supabase, RevenueCat, LLM API
• Tools: Figma, Visual Studio Code, GitHub

Key Requirements

• Chat: Smooth, adaptive, and judgment-free conversation
• Vocabulary: Quick saving and review of words
• Onboarding: Clear guidance for first-time users

Release Journey

At first, I planned a simple chat app with Luna's image as the main feature. To move faster, I started with sketches on paper. But after the App Store rejected it for being too similar to a website, I refocused on enhancing the chat experience and adding personalized features.

I designed chat scenarios to give learners a more natural way to engage with Japanese culture. Users can swipe to go back between screens, and buttons are concentrated at the bottom to make the app easier to navigate.

I enhanced engagement with Duolingo-inspired gamification, using icons for clarity, a dynamic progress graph, and daily random chat topics and encouragement from Luna.

On the settings screen, users can set their images and choose chat formats. I wrote over 130 prompts for Luna to enable natural, patient, and adaptive Japanese conversation practice.

Icon

I first used a moon icon for Luna, but it felt plain. Adding silver-like textures made it into the unique icon.

After Launch

Based on user feedback after launch, I implemented several UX improvements in the Chat and Settings screens.

On the chat screen, I displayed the number of chats remaining and showed the AI's “thinking” status with a loading icon. And I moved the report function to an icon and enabled long-press for copying, which improved MAU by 40%.

When the chat limit is reached, an alert guides users to the Revenue screen, resulting in over a 50% increase in purchases and trial conversions.

On the settings screen, I clarified the chat format language titles and changed the selection from radio buttons to a dropdown menu. I also made the reminder on/off status immediately visible by switching it to a toggle button.

User Flow

Using RevenueCat data showing most purchases happen shortly after first launch, I designed LunaTalk's user journey so trial or purchase can occur within five minutes.

The Solution

LunaTalk was developed with seven core screens. App Store/Google Play screenshots were optimized using ASO and SEO principles, achieving over 1,000 impressions per month without any paid advertising. Try it for free!

The Result

• Outcomes: Daily active users continue to rise, showing strong engagement. ASO improvements increased CVR by up to 50%, and MRR keeps growing. Careful coding has maintained a 0% crash rate.
• Solved Challenge: Delivered natural, judgment-free Japanese conversation practice. Created a personalized, adaptive AI that matches each learner's pace and preferences.
• Impact: Users can confidently practice Japanese, making learning fun, consistent, and rewarding.

Feedback & Iteration

• Feedback Collected: User feedback was gathered via email, in-app anonymous reporting, and online meetings. Users frequently used the chat feature and wanted to copy unknown words by long-pressing messages. Initially, long-press triggered the report function, creating a friction point.
• Insights Gained: Key insights included the need for easier vocabulary saving and more diverse conversation topics. Short daily practice sessions were most effective in keeping learners motivated.
• Iterative Improvements: I moved the report function to an icon and enabled long-press for copying, which improved MAU by 40%. I also enhanced the AI's contextual understanding and refined the vocabulary saving UI for quicker access and review.

Learnings

• Key Insights: Learned how to optimize LLM responses for real language learning and the importance of intuitive UI/UX to build user trust.
• Challenges Overcome: Integrated a robust LLM with real-time chat, keeping low latency and natural dialogue across skill levels.
• Next Steps: I plan to add voice features for real-time pronunciation feedback to boost user satisfaction, refine adaptive learning paths, and expand LunaTalk to other languages.